Advice given to someone asking about AI Consulting: I don't think an ML background is required to be successful in AI consulting, but obviously helps. I think the biggest "skill" learned in ML is how to successfully do feedback loops in a system. In an ML system, this typically

🤫 the PMs don’t know about “claude time” they really think it’s going to take 5-9 weeks to roll out this change don’t inform them otherwise

one of my favorite experiences in my consulting is seeing the teams' excitement after we fix their observability / monitoring issues teams go from minimal capabilities to FAANG-grade monitoring and alerts being able to detect but also be proactively notified when issues occur

slowly becoming convinced async in python was a mistake not because it is bad, but it invites risk for those who aren’t familiar with concurrency and is easy to mess up

you said you were locking in anon, where is the proof? didn't see you in the GH mines

never forget what they took from us

waclaude sitting scanners ready after all... why not?

Rule of thumb for solving a hard problem with LLMs: start with the best in class model. Prove that the problem is tractable once you have the solution where it needs to be, then you can approach optimization around cost/latency the reverse is asking for pain/failure

your job is to find a forever game that you want to spend your working hours playing.

for all you whiney "dOn'T rePorT tHe vuLn iF yoU doN't haVe a pAtch" devs you're in luck i'm fixing this

compaction bug fixed!! huge quality of life improvement

they all come around eventually some learn the lesson sooner than others, but they all learn

“Layoffs due to AI” if you’re bad at executing, just say that

ffmpeg drama summarized

the funniest outcome of Anthropics safety research: we find a way to hold a computer accountable

not every vulnerability needs to be fixed you can "accept risk" and move on

“so you think responsible disclosure is a waste of time” “that’s right Dave” “And that the responsible thing is to just publicly disclose it and avoid all the drama” “that’s right”

“but the volume is fatiguing” i find there is this weird stigma of software vendors where they fear a disclosed CVE and make a big stink about it. the entire point is to inform the community. there is no obligation to fix it. stop treating it like the mark of the beast and

this is the wrong take imo calling a completely valid security report, “slop” is harmful to the community. ��But it’s a hobby codec” the vulnerability exists, it has been disclosed. The community is now informed. everyone wins. put the “accept risk” in the bag

if you thought the software craft declined with the offshoring era just wait till you see the hollowed out AI, no engineers, era

TSA but for websites you want to publish your slop app in my CIDR block, I think not… put the source in the scanner